Development of a Raspberry Pi Drowsiness Detection System based on Histogram of Oriented Gradient (HOG) Algorithm and Eye Aspect Ratio (EAR) Formula

• Main Author: Francis Xavier, Sam Daniel
• Format: Final Year Project
• Language: English
• Published: IRC 2020
• Subjects: Q Science (General)
• Online Access: http://utpedia.utp.edu.my/21699/1/24457_Sam%20Daniel%20a_l%20Francis%20Xavier.pdf; http://utpedia.utp.edu.my/21699/

Driver’s drowsiness is one of the major contributing factors towards the increasing number of accidents in the world. Although there are numerous studies to develop a drowsiness detector system based on driver’s physiological and vehicle-based measures, there are only a few researches conducted to develop a drowsiness detector based on driver’s behavioral measures such as yawning, eye closure or head nodding patterns. Also, another motivation for this research is that most of the drowsiness detection system are only implemented to continental cars, but not for local or inexpensive cars. This is due to the systems’ high-power usage nature, usage of expensive technologies and difficulties in integrating the detection system into all vehicles’ system. Therefore, a Raspberry Pi drowsiness detection system based on Histogram of Oriented Gradient (HOG) algorithm and Eye Aspect Ratio (EAR) formula has been developed in this research. The proposed system in this paper constantly acquires the image of the driver’s face through the attached front camera, conducts two phases of image analyzation, which are detection of facial structure and localization of the eyes and further monitor the changes in the eye aspect ratio values acquired from the image analyzation to detect whether the driver is drowsy or not. The proposed system has also achieved low power consumption and high quality of effectiveness and accuracy in detecting drowsiness. A total number of nine experiments such as placement on different angles and detecting different face positons were conducted to assess the effectiveness and accuracy of the developed system. Hence, efficient analyzation and lower battery usage are assured in the usage of the system. As for further enhancement, yawning monitorization can be integrated with the system for better analyzation and detection. As overall, this paper proposes the development of a cost and power saving and effective drowsiness detection system by implementing EAR formula and HOG algorithm, which would be easily fixed and utilized in all type of four-wheeled vehicles.